Remote Sensing Super-Resolution Reconstruction Based on Improved Deformable Convolution

This paper presents a deformable convolutional alignment algorithm guided by optical flow for super-resolution reconstruction of remote sensing images. The accuracy of reconstruction in remote sensing image super-resolution heavily relies on the alignment process of multi-frame images. Deformable alignment, utilizing second-order grid propagation and flow guidance, has been proven effective in fully exploiting information from the entire input video. However, the long-distance propagation of the alignment structure can result in error accumulation and propagation. To address this issue, a multi-scale encoding module is introduced to mitigate error accumulation and transmission during optical flow alignment, thereby enhancing the model's performance in handling complex detailed textures. Furthermore, to address the blurring problem of small targets and complex textures in remote sensing images, a fully convolutional masked autoencoder is employed to improve feature extraction capabilities, increase the depth of feature extraction, reduce information loss, and ensure information integrity. Experimental results demonstrate the significant performance improvement of the proposed algorithm in the super-resolution reconstruction task for remote sensing images, highlighting its potential for practical applications. This study offers an effective solution for the super-resolution reconstruction task in the field of remote sensing image processing.